Add 'opt_sharding' option for Variables

[samanklesaria]

This PR lets the 'opt_sharding' metadata key shard Variables' state differently from their optimizer state.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[gemini-code-assist]

Summary of Changes

Hello @samanklesaria, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request enhances Flax NNX's sharding capabilities by introducing a dedicated 'opt_sharding' option. This allows for independent sharding configurations for model variables and their associated optimizer states, providing greater flexibility for users to optimize memory and computation in distributed training environments. The change is supported by updated documentation and a new test case.

Highlights

• Separate Optimizer Sharding: Introduced the 'opt_sharding' metadata key, allowing users to specify a sharding configuration for an optimizer's state that is distinct from the variable's own sharding.
• Documentation Update: The 'flax_gspmd' guide (both '.ipynb' and '.md' versions) has been updated to inform users about the new 'opt_sharding' field and its purpose.
• Core Logic Implementation: Modified the 'to_opt_state' function in 'flax/nnx/training/optimizer.py' to correctly apply the 'opt_sharding' metadata when converting a 'Variable' to an 'OptVariable'.
• Test Coverage: Added a new test case, 'test_separate_opt_sharding', to 'tests/nnx/spmd_test.py' to ensure the correct behavior and application of the 'opt_sharding' functionality.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• docs_nnx/guides/flax_gspmd.ipynb
  • Clarified that 'sharding_names' also applies to 'nnx.Optimizer' and introduced 'opt_sharding' for separate optimizer sharding.
• docs_nnx/guides/flax_gspmd.md
  • Added information about 'opt_sharding' to specify distinct sharding for optimizer states.
• flax/nnx/training/optimizer.py
  • Modified '_to_opt_state' to check for 'opt_sharding' in variable metadata and use it as 'sharding_names' for the 'OptVariable' if present.
• tests/nnx/spmd_test.py
  • Added 'test_separate_opt_sharding' to verify that 'opt_sharding' correctly applies a different sharding to the optimizer's state compared to the variable's state.

Activity

• The author samanklesaria introduced a new feature to allow separate sharding for variables and their optimizer states.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces the opt_sharding metadata key for nnx.Variable. This is a useful feature that allows specifying a different sharding for an optimizer's state compared to the variable's own state. The implementation in flax/nnx/training/optimizer.py correctly handles this new key by prioritizing it for the optimizer state's sharding. The documentation has been updated accordingly in flax_gspmd.ipynb and flax_gspmd.md, and a new test case in spmd_test.py verifies the functionality. The changes are well-implemented. I have one minor suggestion to make the optimizer state's metadata cleaner.